This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

IEEE TRANSACTIONS ON SMART GRID

Stability Analysis of Networked Control in

Smart Grids
Abhinav Kumar Singh, Student Member, IEEE, Ravindra Singh, Member, IEEE, and Bikash C. Pal, Fellow, IEEE

AbstractA suitable networked control scheme and its stability

analysis framework have been developed for controlling inherent
electromechanical oscillatory dynamics observed in power systems. It is assumed that the feedback signals are obtained at
locations away from the controller/actuator and transmitted over
a communication network with the help of phasor measurement
units (PMUs). Within the generic framework of networked control system (NCS), the evolution of power system dynamics and
associated control actions through a communication network
have been modeled as a hybrid system. The data delivery rate has
been modeled as a stochastic process. The closed-loop stability
analysis framework has considered the limiting probability of data
dropout in computing the stability margin. The contribution is
in quantifying allowable data-dropout limit for a specified closed
loop performance. The research findings are useful in specifying
the requirement of communication infrastructure and protocol
for operating future smart grids.
Index TermsCyber physical energy network, damping control,
Kalman filtering, linear quadratic Gaussian (LQG), networked
control systems (NCS), packet-drop, phasor measurement units
(PMUs), smart grid, stochastic stability, thyristor controlled series
capacitor (TCSC).

I. INTRODUCTION

LECTRICAL energy systems all over the world have

been undergoing unprecedented level of transitions
in recent times. The transitions are primarily driven by the
requirement to renew the aging electricity infrastructure; to
integrate low-carbon intermittent energy sources; to meet
increased power consumption and new forms of demands such
as electric transport, smart home and business; and to ensure
the security of supply. Several enabling technologies such as
power electronics, wide area measurements (WAMS) and the
information and communication technology (ICT) are increasingly getting deployed in the system to deliver the requirement.
Such deployments increase the complexity and challenges of
the system operation and the electricity grid of the 21st century
has to embrace these challenges through planning, design,
operation and control. In the operation time-scale ensuring the
security and stability of the system are important tasks. The
WAMS and ICT offer very effective means of delivering these
objectives ([1] and [2]).

Manuscript received November 14, 2013; revised February 19, 2014 and
March 24, 2014; accepted March 27, 2014. This work was supported by EPSRC,
U.K., under Grant EP/G066477/1. Paper no. TSG-00854-2013.
A. K. Singh and B. C. Pal are with the Control and Power Group, Department
of Electrical and Electronic Engineering, Imperial College London, SW7 2BT,
London (e-mail: a.singh11@imperial.ac.uk; b.pal@imperial.ac.uk).
R. Singh is with the ABB Corporate Research Centre, Rayleigh, NC, USA
(e-mail: singh.ravindra@gmail.com).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TSG.2014.2314494

The networked control system (NCS) approach utilizing

modern measurements and communications is very appropriate
in this context. In NCS the control loops are closed through
a real-time network where the control and feedback signals
are exchanged among the systems components in the form
of packets of data and information [5]. It enables execution
from long distance by connecting cyberspace to physical
space. It has been successfully applied in other technology
areas such as space and terrestrial exploration, aircraft, automobiles, factory automation and industrial process control.
The Packet-switching based communication networks are the
most widely adopted systems for fast, economic and stable
data transfer over both large and small distances through dynamic path allocation. They are in contrast to the traditional
circuit-switching based networks in which a dedicated link
is established between the sending and the receiving ends.
Circuit switching is not only inefficient and costlier than
packet-switching, but also the link failure rate increases for
large transmission distances, and the failure cannot be dynamically corrected, unlike packet-switching [3], [4]. This is the
reason that most of the current research in NCS is based on
packet-switching technology. The NCS offers many advantages
over the traditional control architectures, including low cost
of installation, ease of maintenance and greater flexibility.
They also suffer from some problems such as packet-dropout,
network induced delays and packet-disordering [6]. These
factors can possibly degrade the performance of the control of
important power system dynamics such as system wide electromechanical oscillations, more commonly known as inter-area
oscillations of the system [7]. In the context of interconnected
power systems, the control of oscillatory stability is very time
critical as uncontrolled oscillations in the past have led to
several power blackouts [8]. Therefore one needs to analyze
these factors thoroughly for assessing the suitability of the NCS
approach to wide area control of power systems.
Over the past decade, substantial research has been undertaken to model NCS and study the effect of packet-dropout
and time delays on the control design and the stability of the
NCS ([5], [6], [9] and [10]), but this research is not reflected in
the power system literature. In most of the literatures relating
to power systems, it is assumed that the transmission of signals to and from the central control unit occurs over an ideal,
lossless and delay-free communication network. A few exceptions to this are [11], [12] and [13]. In [11] the effect of network induced time-delays has been considered using a WAMS
based state-feedback control methodology. In [12] an estimation of distribution algorithm (EDA) based speed control of networked DC motor system has been studied; and in [13] the effect of communication-bandwidth constraints on the stability of

1949-3053 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
2

IEEE TRANSACTIONS ON SMART GRID

WAMS based power system control has been studied. But all
these papers have other limitations. For instance, in [11] it is not
explained how the various system states (such as the rotor angle,
rotor velocity and transient voltages) are estimated before using
them for state-feedback; and also the power system model considered in the paper is too simplistic to represent actual power
system dynamics. In [12] only a local network based control of
a single dc-motor system is considered instead of considering
the networked control of a complete power-system. In [13], the
chief problems associated with networked-control, which are
packet-loss and delay, are not considered.
This paper has made an attempt to address the aforementioned limitations by analyzing the effects of packet-dropout
on the oscillatory stability response of a networked controlled
smart grid (NCSG). A Kalman filtering and linear quadratic
Gaussian (LQG) based optimal control scheme for damping oscillatory stability of the NCSG has been adopted in the paper.
Specifically, the contributions of the paper may be summarized
as follows:
A detailed characterization of packet transmission process
and the probability of packet loss have been considered in
the framework of NCS for power system control.
A practical output-feedback methodology has been used
for control (instead of state-feedback), and the signals
which are required to be transmitted to the control unit
are measureable line-power signals. Also, a detailed and
realistic sub-transient power system model has been used.
The LQG control scheme used in the paper is an optimal
control scheme for a networked system in which packet
loss is considered. Moreover, the control scheme can be
easily integrated with the WAMS or flexible AC transmission system (FACTS) devices already present in the
system, and any extra investment is not required.
A rigorous model for a NCSG is presented in Section II.
Section III presents LMI based stability analysis of the developed NCSG, and derives the probability threshold of the
packet-dropout rate while guaranteeing specified level of
damping of the NCSG. A case study of a representative 68-bus
New-England/New-York inter-connected NCSG model has
been presented in Section IV. In the case study, the inter-area
oscillations in the power system are controlled using feedback
signals which are transmitted over a communication network.
Section V concludes the paper.

II. NCSG MODELING WITH OUTPUT-FEEDBACK

A block diagram of the output-feedback controlled NCSG is
shown in Fig. 1. The model is described as a hybrid continuousdiscrete system in which the power system is the continuous,
while the networked-controller is the discrete part.
In Fig. 1, the block Power System represents the open-loop
power system whose oscillatory dynamics is to be controlled.
To this effect, real-power deviations in some of the lines are
measured in real-time using current transformers (CTs) and potential transformers (PTs), and represented by
in the block
diagram. These are then sampled at the sampling rate of the
communication network using digital devices such as phasor
measurement units (PMUs) and intelligent electronic devices

Fig. 1. A reduced model of the NCSG.

Fig. 2. Details of the power system in the NCSG.

(IEDs) and then sent over the communication network as discrete data-packets,
. User datagram protocol (UDP) is used
for packet transmission, and packet-loss occurs during transmission. The final data which is received at the control unit after
packet loss is given by
. The control unit consists of a
LQG controller, which is a combination of a Kalman filter and
a linear quadratic regulator (LQR). Kalman filter uses the reduced-power system model and the output data-packets arriving
at the controller,
, to estimate the states,
. The state
estimates are then multiplied by the LQR gain to produce the
control signals
, which are then sent over the communication network to the actuators. The packets arrive at discrete to
analog converters (DACs), which are zero order hold devices
and convert the discrete control signals after packet-loss,
,
. These continuous siginto continuous control signals,
nals control the actuators, which are the FACTS devices, more
commonly known as FACTS controllers. The inputs
to the
power system are the percentage compensations provided by
the FACTS controllers to control the power-flow in the lines
on which the FACTS controllers are installed. All the variables
in the model have been expressed in per unit (p.u.), except the
time variables which are expressed in s. A detailed description
of each component of the NCSG model is presented as follows.
A. Power System
An interconnected power system is represented through
important components such as the generators, their excitation
systems, power system stabilizers (PSS), FACTS controllers
such as a thyristor controlled series capacitor (TCSC), loads
and transmission network [15] as shown in Fig. 2.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
SINGH et al.: STABILITY ANALYSIS OF NETWORKED CONTROL IN SMART GRIDS

The dynamics of the system is modeled using a set of nonlinear differential and algebraic equations (DAEs) ([14] and
[15]). The state space representation of the system is obtained
through linearization of the DAEs around an initial operating
point. The order of the system is generally reduced to speed
up the controller design algorithm. On applying a standard balanced model reduction given in [16], only the unstable and/or
poorly damped electromechanical modes of the power system
are retained in the reduced model. The reduced model is written
as:
(1)
(2)
,
and
are the reduced state
space matrices and
,
and
are the vectors
of reduced-state variables, inputs and outputs, respectively. It
should be noted that after balanced reduction of the full model,
only the state variables and the state matrices get reduced in
order; the inputs and the outputs remain same as in the original full model. Also, out of the various possible measurable outputs (which are the line-powers in the context of NCSG), only
those outputs are selected in which have high observability of
the unstable and/or poorly damped electromechanical modes of
the power system.
B. Sensors and Actuators
The sensors (in the context of NCSG, they are CTs, PTs and
PMUs) send the feedback signals to the controller over the communication network at a regular interval of , which is the
sampling period of the communication network. The discrete to
analog converters (DACs) convert the discrete control signals
after packet-loss into continuous control signals. The DACs are
event-driven zero-order-hold (ZOH) devices, each one of which
holds the input to the power system in a given cycle. In the next
cycle it holds its previous value if there is no new input due to
packet drop, otherwise it holds the new input. The outputs of the
DACs control the FACTS controllers, which are the actuators;
and the inputs
to the power system are the percentage compensations provided by the FACTS controllers. For the
th
time cycle, (1) reduces to:
(3)
Solving (3) with initial condition
constant input
[17], we get:

and a

(4)
Denoting
as ,
as
(where
is
the uncertain input after packet dropout while
is the white
as ,
as , and the
Gaussian noise at actuators),
white Gaussian noise at the sensors as , we get:
(5)
C. Communication Protocols and Packet Delays and Dropouts
In the model design process two classes of communication
protocols have been considered. In transmission control pro-

Fig. 3. Markov chain for Gilbert process.

tocol (TCP)-like protocols the acknowledgments that the receiver received the packets are sent back to the sender, while
in user datagram protocol (UDP)-like protocols they are not
sent. In TCP-like case, unlike in the UDP-like case, the lost
packets can be re-sent because of the availability of the acknowledgments. So the separation principle, as explained in
[18], holds only in the case of TCP-like protocols, and hence the
controller and the estimator can be designed independently [19].
In UDP-like case no known optimal regulator exists and one can
design a suboptimal solution based on a Kalman-like estimator
and a LQG-like state feedback controller, as shown in Fig. 1. Although UDP-like protocol results in a sub-optimal solution, it is
preferred over a TCP-like protocol as it may be extremely difficult to both analyze and implement a TCP-like control scheme
[19]. In this paper, a UDP-like scheme has been used. The time
delays and dropouts of packets have been modeled such that a
packet is assumed to be lost, unless its time-delay is less than
the sampling interval of the system. This fact is one of the factors while deciding the sampling duration, the other factor being
the type of control needed, as explained in Section IV-B1. If a
packet is lost, the output of the receiver is held at the last successfully received packet.
The packet loss over the network usually follows a random
process. In the present analysis an independent Bernoulli
process has been used to model the packet loss [19]. The input
at the actuator and
at the estimator are modeled as:
(6)
is a stationary diagonal binary
random matrix, in which the value of
is equal to one with a
probability
, indicating that the th component of
is deliv,
ered; while its value is equal to zero with a probability
indicating that the component is lost.
Remark: The assumption of an independent Bernoulli packet
loss model is not valid when the communication channel is congested. In a congested channel the packet loss occurs in bursts,
and follows a two-state Markov chain model, also known as
Gilbert model [20]. Fig. 3 shows this model, where 1 represents the state of packet delivery and 0 represents the state of
packet loss; and the probability of transition from state 0 to
state 1 is and the probability of transition from state 1 to
state 0 is . When is equal to
, then this model reduces to Bernoulli model.
The drawback of using Gilbert model in stability analysis
is that this model is not a memory-less model, which means
that the probability of packet delivery depends on the current
channel state, and it fluctuates between and
depending
on whether the current channel state is 0 or 1, respectively.
Mathematical representation of such a fluctuating probability of
packet delivery becomes practically infeasible. A practical alter-

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
4

IEEE TRANSACTIONS ON SMART GRID

native for approximating Gilbert model with Bernoulli model

can be to set the communication channels probability of packet
delivery as if
, and as
if
. Thus,
the approximated Bernoulli model represents the worst case scenario of packet delivery performance given by Gilbert model, as
the smaller probability of the two possible packet delivery probabilities from Gilbert model is assumed to be the constant packet
delivery probability in the approximated Bernoulli model.
D. Controller
For an open-loop LTI system given by (5), whose input
defined by (6), the quadratic cost function is given by:

is

(7)
where is the number of samples, is the expectation value,
denotes the transpose of a vector or a matrix,
is a positive definite matrix,
is a positive semi-definite matrix and
it is assumed that the full state information of the LTI system
is available (we get this information from the state-estimator).
Minimizing with respect to
results in the following forward Riccati-like difference equation, as exlained in [21]:

(11)
2) Estimation step:
(12)
(13)
(14)
For the kth sample,
is the predicted state,
is the
is the estimated
predicted-state-error covariance matrix,
is the estimated-state-error covariance matrix,
state,
is the covariance matrix of
,
is the covariance matrix of
,
is the Kalman gain. The equations are valid iff
is
is controllable. In (10) the estimator
observable and
takes the closed loop state-space matrix
as
as it can at best have an estimate of the packet dropout rate of
the network because it does not receive the acknowledgments
of the control packets it sends out to the power system.
III. CLOSED-LOOP STABILITY AND DAMPING RESPONSE
The closed loop model of the NCSG can be summarized as
follows, using (5)(14) and rewriting
as :
(15)
(16)

(8)

(17)
is a positive definite matrix, say an identity matrix,
where
of the same order as and
is the expectation value of
(subscript is removed in
as
is stationary). If we obtain a steady state solution
for (8) as
then
the LTI open-loop system is stabilizable in mean-square sense,
provided the pair
is controllable; the pair
is
, and perfect state inforobservable, where
mation is received at every sample. The infinite horizon control
policy for such a system is a state feedback policy, given by:
(9)
where

is the estimated state, and

A steady state solution for

in (13), and hence for , may
or may not exist for given
and , even if the conditions
for the existence of steady state solution for a standard Kalman
filter hold; but a steady state estimate
for the Kalman
gain may be obtained by iteratively solving (11), (13) and (14)
after substituting
with its expected value
. This is the
sub-optimal Kalman gain which is used for deriving the condition for mean square stability and adequate damping of the
developed NCSG. Writing (15)(17) in composite form, after
replacing
with its steady state estimate , we get:

is the LQG gain.

E. Estimator

(18)

The controller uses the output from the state estimator to generate the control command which is sent over the network to
the actuator in the power system. The estimator uses the information vector, which consists of the control command and the
intermittent plant output delivered to the estimator via the network, to generate a best estimate of the state of the system. It
was shown in [22] that even in the case of intermittent observations, Kalman filter is still the best linear estimator for LTI
systems with stationary Gaussian noise processes, provided that
only time update is performed when a measurement packet is
dropped. When a measurement is received, both the time and
measurement update steps are performed. The filtering equations for such a closed-loop system, using (9), are:
1) Prediction step:
(10)

The presence of
and
in (18) makes it a jump linear
system (JLS): a system whose state matrices vary randomly with
and
. The framework of a JLS and its stability analysis
are described in [23], [24] and [25]. A brief overview of the
criterion for the stability and the damping in mean square sense
of the NCSG has been presented in the next section.
A. Stability Analysis Framework of a Jump Linear System
. Let
Let be a set of all the subsets of
be a set of indices of all those input delivery indicators whose
values are one, i.e.,
. E.g., for a
2 input system
, can either be (both the inputs failed
to deliver), or {1} (only 1st input delivered), or {2} (only 2nd

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
SINGH et al.: STABILITY ANALYSIS OF NETWORKED CONTROL IN SMART GRIDS

input delivered), or {1,2} (both the inputs delivered). Similarly,

let
be a set of indices of successful output delivery
indicators. As each input delivery indicator
has two modes
(0 or 1) and
are in total,
has
modes. Any mode
of
is expressed as
,
, where
is a
diagonal matrix whose
th element is 1 if
, else it is
0. Similarly,
has
modes, and any mode is expressed
as
,
, where
is a
diagonal matrix
whose
th element is 1 if
, else it is 0. The probability
distributions of
and
are given by:
(19)
(20)
As
in (18) is a function of
be re-expressed as
in (21):

and

, it may

(21)
As the value of
depends on the values of and , it
can take any value in a given sample out of the possible
values, with a corresponding overall probability distribution
. The NCSG in (18) is said to be the mean-square
stable if

, starting with any state

The mean-square stability of (18) is checked using the following methods:

1) Linear Matrix Inequalities: The criterion for the stability
of discrete-time JLS in [24] is applied to obtain the condition
for the mean-square stability of the NCSG in (21). The satisfaction of the criterion requires the existence of positive definite
matrices
,
,
, such that:

(22)
There are in total
LMIs in (22) and hence this method may
be practically used only when the inputs and the outputs to the
system are small in number. For a system with relatively larger
number of inputs and outputs, the following stochastic stability
analysis for JLS has been used.
2) Stochastic Eigenvalue Analysis:
Lemma 1: Consider the closed-loop NCSG described by
(18). For given packet delivery probabilities (PDPs)
and
,
the NCSG is mean-square stable iff
,
is Schur stable (all of its eigenvalues are within unity circle).
Various elements of
are defined, explained and
followed by the proof of this lemma in Appendix A.
B. Analysis of Damping Response
The concept of -stability [26] has been used to study the
adequate damping response of the developed NCSG. This is
very practical and useful in the context of power oscillation
damping. If
is a sub-region of the complex left half plane,
and all the closed loop poles of a dynamical system
lie

Fig. 4.

-stability region for damping control of a continuous system.

in

, then the system and its state transition matrix are called
-stable. When is the entire left-half plane, then -stability
reduces to asymptotic stability. For damping control analysis,
the -region of interest is
of complex numbers
s.t.
(Fig. 4). Thus a specified and required damping
of inter-area modes becomes an important criterion for NCS design and analysis. Confining the closed loop poles of the system
to this region ensures a minimum damping ratio
.
This in turn bounds the decay rate and the settling time for the
corresponding oscillatory inter-area modes of the system. Power
systems usually require an operating constraint that all the disturbances in the system should settle to less than a fixed percent
(usually 15%) of the maximum overshoot within a few seconds
(usually 1015 s) of the start of the disturbance to the system
[28]. As the inter-area modes usually lie in the frequency range
0.21.0 Hz, they have longer settling times and lower decay
rates than other modes. In this paper, the closed-loop inter-area
modes have been considered to be damped to more than ten percent.
Lemma 2: The closed-loop NCSG in (18) is expected to have
all of its equivalent continuous-time poles with damping ratios
iff there exist positive definite matrices
,
,
, such that:

(23)

where,

is the natural logarithm of a square matrix and is the sampling-period of the NCSG.
Appendix B has the proof of this lemma.
Remark: If all the channels have same characteristics, then
their PDPs become equal to each other
.
The marginal packet delivery probability (MPDP), such that the
NCSG remains properly damped
, is given by
.
C. Physical Significance of Lemma 1 and Lemma 2 in the
Context of Power Systems
The physical meaning of the mathematical results given by
Lemma 1 and Lemma 2 will be better understood using the
concepts of observability and controllability. As mentioned in
Section II-A, the output measurements have high observability

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
6

IEEE TRANSACTIONS ON SMART GRID

of the unstable and/or poorly damped electromechanical modes

of the power system. The LQG controller requires the knowledge of these measurements and the state matrices in order to
correctly estimate the states, which are then multiplied by the
LQR gain to get the control input for the power system. The
LQG controller requires the knowledge of these measurements
and the state matrices in order to correctly estimate the states,
which are then multiplied with the LQR gain to get the control
input for the power system. The closed loop system is properly stabilized and damped, provided the packet delivery rate is
100%. The decrease in packet delivery rate from 100% results in
the loss of the output measurements in the communication network. The measurements which finally arrive at the controller
after packet loss have an overall decrease in their observability
for a given period of time, and the controller estimates the states
with decreased accuracy. For a packet delivery rate of zero percent, none of the measurements arrive at the LQG controller, and
thus the observability is zero, and controller cannot estimate the
states at all.
This concept of probabilistic observability will be better
understood with an example. For example, if there are two
measurements which are sent over the network then there
are four possibilities in a given time sample: 1) none of the
measurements arrive, 2) only first measurement arrives, 3) only
second measurement arrives, and 4) both the measurements
arrive. Each of these four possibilities has a probability associated with it depending on the packet delivery rates of the
two communication channels. The overall observability of the
arriving measurements depends on these four probabilities, and
is thus a probabilistic quantity in itself. For measurements,
there are
probabilities, and the overall observability will
depend on all of these probabilities. Similar analogy applies for
the controllability of the power system by the control inputs
sent over the communication network, and thus the overall
controllability is also a probabilistic quantity. The stability
and the -stability of the closed loop system depend on these
probabilistic observability and controllability, and are written
in mathematical forms as Lemma 1 and Lemma 2, respectively.
IV. CASE STUDY: 16-MACHINE 5-AREA NCSG
A. System Description
A 16-machine, 68 bus model test system, shown in Fig. 5,
has been used for the case study. This is a reduced order equivalent of the interconnected New England test system (NETS)
and New York power system (NYPS) of the 1970s. NETS
and NYPS are represented by a group of generators, while
the power import from each of the three other neighboring
areas are approximated by equivalent generator models (G14
to G16). NYPS needed to import around 1.5 GW from Area 5,
for which a TCSC was installed on the 1850 tie-line. It was
needed to dynamically control the percentage compensation of
the TCSC to control the reactance of the tie-line. A detailed
system description is available in [28] which is used to simulate
in MATLAB SIMULINK.
B. Simulation Results and Discussion
1) Operating Condition 1 (Base Case): Total Tieline Flow
Between NETS and
, No Line Outages:

Fig. 5. Line diagram of the 16-machine, 68-bus, 5-area NCSG.

TABLE I
NORMALIZED RESIDUES OF THE ACTIVE POWER-FLOWS IN THE 3 MODES

For the first case of system operation, the damping and the
frequency of the three poorly damped modes of the linearized
system were computed. The result is displayed in Table I. The
open loop system response confirmed that the other electromechanical modes including one inter-area mode of the system
settled in less than 10 seconds and hence they have been left
from the consideration of providing additional damping. The
remote feedback signals were chosen based on modal observability analysis [29] for various active line power signals. For
the fixed location of actuator, the modal controllability does not
change so modal residue is related to modal observability by a
scale factor. Table I gives the normalized residues of top 3 active power flows in the 3 inter-area modes.
The signals
(having highest residues for modes 1 and
3) and
(having highest residue for mode 2) have been
selected as output signals. With these two signals as output
and
, (the control signal of the TCSC) as the input, the
open-loop system was linearized to find the state space matrices. System order was reduced (Section II-A) to the lowest
possible order such that the reduced system still remained a
very good approximation of the full system in the frequency
range of 0.21.0 Hz; and thus a reduced seventh order system
was obtained, as shown in Fig. 6.
Remark: Deciding the Sampling Period: The controller
needs to observe the system in the range of 0.21.0 Hz; hence
the minimum required sampling frequency is 2.0 Hz (i.e.,
a maximum allowed sampling period of 0.5 s) according to
the Nyquist-Shannon sampling theorem. This upper limit of
sampling period is also the threshold requirement for average
time delay of the packets, as if the average delay is more than
this upper limit, then the packet loss rate will be very high and
the network would not support the communication needs of

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
SINGH et al.: STABILITY ANALYSIS OF NETWORKED CONTROL IN SMART GRIDS

Fig. 6. Frequency response of the full vs. the reduced system.

the system. In this paper, a conservative sampling period of

was assumed.
The packet loss in the path of the input and output signals was
modeled as a Bernoullis process. So,
, while
, in (6). The steady state controller gain for
the reduced system was found using the results of Section II-D
and the modified Kalman filter was modeled using the principle
described in Section II-E. Simulation was started and after one
second a disturbance was created in the NCSG model by a threephase fault and immediate outage of one of the tie lines between
buses 5354.
The open loop system was a minimum-phase system (i.e.,
all of its modes were stable); thus it was required to check only
the damping response of the system for various packet drop
rates. For ,
, and
has two modes,
and
. For ,
and
,
has four modes viz.
,
,
and
and
. The corresponding jump state matrices
are
,
,
,
,
,
,
and
, and their probabilities of occurrences are
,
,
,
,
,
,
, and
, respectively. Using these parameters, eight pairs of
LMIs in (23) were obtained. was taken as 84.3 degrees corresponding to 10% damping line, as shown in Fig. 4. Assuming
same network characteristics for all the network-channels,
i.e.,
, the feasibility of the LMIs
using LMI toolbox in
was checked for various values of
MATLAB. The toolbox returned a minimum feasible value of
, i.e., the LMIs were feasible for
.
Fig. 7 shows the rotor slip response for G16 at the marginal
PDP of
. System response using a classical damping
controller [30] (assuming a perfect communication link in its
control loop, with infinite sampling rate and zero packet loss)
has also been shown for comparison. The details of the classical
controller are available in the Appendix C.
The rotor-slip response of G16 following the disturbance has
also been displayed for four other values of
as shown in
Fig. 8.
Remark: It should be understood that it is not the sole purpose
of Fig. 7 (and subsequent figures) to show that the performance
of the networked controller is better than the classical damping
controller. Rather, its another important purpose is to show that
the performance of the networked control with communication
packet-dropout, even with marginal PDP, is comparable to the
performance of classical control in which an ideal, lossless and

Fig. 7. Rotor-slip response for G16 at operating point 1.

Fig. 8. Rotor-slip response for various packet delivery probabilities (PDPs).

Fig. 9. Classical vs. networked control, with assumption of ideal network conditions.

delay-free communication network is assumed. Fig. 9 shows the

comparison of the performance of networked control with that
of classical control, when in both the cases an ideal communication network is assumed (that is
). It can be clearly
verified from the figure that the performance of networked control is much better than classical control when ideal network
conditions are assumed for both the cases. Also, a metric which
is used to assess the control effort required by a control method
is the 2-norm of the output from the controller, or
. The
2-norm for the classical control is 0.32 p.u., while for networked
control (with
) it is 0.25 p.u. Thus networked control at 100% packet delivery rate is better than classical control,
and it can damp the oscillations in a smaller amount of time,
even when the control effort required by networked control is
decreased by 22% as compared to the control effort required by
classical control.
2) Operating Condition 2: Total Tieline Flow Between NETS
and
, No Line Outages: In the second operating condition, the open-loop system becomes unstable after
the line-outage, unlike the first operating condition. This is due
to the presence of an unstable mode with negative damping
ratio in the system. Therefore we may apply either the LMI

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
8

IEEE TRANSACTIONS ON SMART GRID

TABLE II
MARGINAL PACKET DELIVERY PROBABILITY VS. OPERATING POINT

Fig. 10. Rotor-slip response for G16 at operating point 2.

Fig. 11. Marginal delivery probability vs. sampling period.

analysis (Section III-A1) or the stochastic eigen value analysis (Section III-A2) to find the marginal PDP which can ensure closed-loop stability of the NCSG. It was found that the
stability of the NCSG under this operating condition was ensured at a marginal PDP of 0.24, while the adequate damping
of the system was ensured at a marginal PDP of 0.87 (using
Section III-B). The slip response of G16 has been shown at both
of these marginal PDPs in Fig. 10.
It is evident in Figs. 7 and 8, for
and PDPs more
than or equal to 0.81, the inter-area modes of the system are
properly damped. Similarly, it may be observed from Fig. 10
that the system in second operating condition is stabilized at the
marginal PDP of 0.24 while adequately damped at the marginal
PDP of 0.87. So the results of the LMI analysis stand verified.
It is also clear from Fig. 7 and Fig. 10 that the performance
of the networked controller is better than the classical damping
controller, at realistic packet delivery qualities that can be easily
delivered by present day telecom networks.
Next, the effect of the sampling period on the marginal delivery probability for -stability is being investigated. Fig. 11
shows the plot of the marginal delivery probability versus sampling period. One can easily infer from Fig. 11 that higher sampling period requires increase in to guarantee feasibility. This
is in line with the expectation that a packet has to be delivered
with higher probability with an increase in sampling time.
The robustness of the NCSG has been checked by obtaining
the probabilities of marginal packet delivery for various operating conditions as listed in Table II. In Table II, serial number 1
(S.No.1) was considered the base case of operation. For each operating condition, the control scheme was required to be updated
to give a corresponding LQG gain and reduced-order state space
matrices for the Kalman filter. The stability performance of the
NCSG was also studied with a constant control scheme, i.e., the
control scheme obtained for the base case was used for all the
operating conditions. Fig. 12 shows the rotor-slip response for
3 operating conditions with such constant control scheme. It is
clear from Table II and Fig. 12 that the NCSG is -stable for

Fig. 12. Rotor-slip response for various operating points at

various operating points, even with a constant control scheme,

at a feasible delivery probability of 0.85.
V. CONCLUSIONS
The effects of introducing a packet-based communication
network in the control loops of a power system have been
analyzed using the generalized model of a networked controlled smart grid. The random loss in the delivery of the
packets has been modeled as a stochastic process. Using the
developed linear matrix inequalities, the lower limit on the
probability of packet delivery has been computed which guarantees specified damping. It was also found that under varying
operating conditions the performance of the NCSG was robust.
The contribution of the paper lies in the fact that it not only
presents a networked control framework for a power system;
but it also presents a formal approach for finding the minimum
network requirements in terms of packet delivery quality, so
that the specified stability and damping margins can be ensured
for any operating condition of a power system. The research
findings show that the NCSG framework has a good potential

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
SINGH et al.: STABILITY ANALYSIS OF NETWORKED CONTROL IN SMART GRIDS

to deliver oscillatory stability margin for the smart electricity

grid network of the 21st century.

Other terms of

are similarly derived as:

APPENDIX A
DETAILS AND DERIVATIONS OF LEMMA 1
The jump system in (18) is mean square stable iff the matrix
is Schur
stable (using the stability result of jump linear systems [25]).
Rewriting

in (21) as

, and approximating

plify the stability analysis, we have

,

,
, and,

to sim,

APPENDIX B
PROOF OF LEMMA 2

The elements of

as

are derived as follows:

and

shown in Fig. 4 is appliThe damping-region of interest

cable only to the continuous-time representation of a dynamic
system; the discrete-time equivalent of this region is a logarithmic spiral and is very difficult to represent using matrices
and LMIs. We therefore consider the continuous-time equivalent of
which is given by (23) as
. Using [27]
we know that a dynamic system
is -stable in the
region
iff
is asymptotically stable. This holds because the eigenvalues of
are
. The eigenvalues
of the Kronecker product of 2 matrices are the product of the
eigenvalues of individual matrices. Hence, the eigenvalues of
are two sets of eigenvalues of , one set rotated by
an angle
and another one by
. All
those eigenvalues of which lie outside
get rotated into
the right half plane in
, and hence
is asymptotically stable iff none of the eigenvalues of lie outside
,
i.e., iff
is adequately damped. So, the asymptotic stability
of
implies -stability of
. As
is a jump-linear mode of the stochastic system in (18) with a
modal probability of
, the matrix
is
also a modal matrix of same probability as
; and the
mean-square stability of
[24] (given by (23)) implies the -stability of
in a mean square sense, i.e., its
electro-mechanical modes are expected to have
.
APPENDIX C
DETAILS OF THE CLASSICAL CONTROLLER
The transfer function for the classical power oscillation
damping (POD) controller used in the text has been evaluated
using the theory and results given in [30], and it is as follows:

REFERENCES
[1] C. Hauser, D. Bakken, and A. Bose, A failure to communicate: Next
generation communication requirements, technologies, architecture for
the electric power grid, IEEE Power Energy Mag., vol. 3, pp. 4755,
Mar.Apr. 2005.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
10

IEEE TRANSACTIONS ON SMART GRID

[2] Z. Xie, G. Manimaran, V. Vittal, A. Phadke, and V. Centeno, An information architecture for future power systems and its reliability analysis, IEEE Trans. Power Syst., vol. 17, pp. 857863, Aug. 2002.
[3] J. E. Flood, Telecommunication Networks, 2nd ed. London, U.K.:
IET, 1997.
[4] T. Y. Feng, A survey of interconnection networks, Computer, vol.
14, no. 12, pp. 1227, Dec. 1981.
[5] G. Walsh, H. Ye, and L. Bushnell, Stability analysis of networked control systems, IEEE Trans. Control Syst. Tech., vol. 10, pp. 438446,
May 2002.
[6] W. Zhang, M. Branicky, and S. Phillips, Stability of networked control
systems, IEEE Control Syst., vol. 21, pp. 8499, Feb. 2001.
[7] J. Paserba, Analysis and control of power system oscillations, in
CIGRE Special Publication 38.01.07, Technical Brochure no. 111,
1996.
[8] G. Andersson, P. Donalek, R. Farmer, N. Hatziargyriou, I. Kamwa,
P. Kundur, N. Martins, J. Paserba, P. Pourbeik, J. Sanchez-Gasca, R.
Schulz, A. Stankovic, C. Taylor, and V. Vittal, Causes of the 2003
major grid blackouts in North America and Europe, recommended
means to improve system dynamic performance, IEEE Trans. Power
Syst., vol. 20, no. 4, pp. 19221928, Nov. 2005.
[9] Y. Tipsuwan and M. Y. Chow, Control methodologies in networked
control systems, Control Eng. Practice, vol. 11, pp. 10991111, Oct.
2003.
[10] J. Hespanha, P. Naghshtabrizi, and Y. Xu, A survey of recent results
in networked control systems, Proc. IEEE, vol. 95, pp. 138162, Jan.
2007.
[11] S. Wang, X. Meng, and T. Chen, Wide-area control of power systems
through delayed network communication, IEEE Trans. Control Syst.
Tech., vol. 20, pp. 495503, Mar. 2012.
[12] H. Li, M. Y. Chow, and Z. Sun, Eda-based speed control of a networked dc motor system with time delays and packet losses, IEEE
Trans. Ind. Electron., vol. 56, pp. 17271735, May 2009.
[13] N. R. Chaudhuri, D. Chakraborty, and B. Chaudhuri, An architecture
for facts controllers to deal with bandwidth-constrained communication, IEEE Trans. Power Del., vol. 26, pp. 188196, Mar. 2011.
[14] K. Padiyar, Power System Dynamics: Stability and Control. Tunbridge Wells, U.K.: Anshan Limited, 2004.
[15] M. A. Pai and P. W. Sauer, Power System Dynamics and Stability.
Englewood Cliffs, NJ, USA: Prentice-Hall, 1998.
[16] M. G. Chiang and R. Y. Safonov, A schur method for balanced model
reduction, IEEE Trans. Autom. Control, vol. AC-34, pp. 729733,
1989.
[17] K. Ogata, Modern Control Engineering, 4th ed. Upper Saddle River,
NJ, USA: Prentice-Hall PTR, 2001.
[18] D. P. Bertsekas, Dynamic Programming and Optimal Control, 1st
ed. Belmont, MA, USA: Athena Scientific, 1995.
[19] B. Sinopoli, L. Schenato, M. Franceschetti, K. Poolla, and S. Sastry,
Optimal linear lqg control over lossy networks without packet acknowledgment, Asian J. Control, vol. 10, pp. 313, Jan. 2008.
[20] X. Yu, J. W. Modestino, and X. Tian, The accuracy of Markov chain
models in predicting packet-loss statistics for a single multiplexer,
IEEE Trans. Inf. Theory, vol. 54, no. 1, pp. 489501, Jan. 2008.
[21] W. D. Koning, Infinite horizon optimal control of linear discrete time
systems with stochastic parameters, Automatica, vol. 18, pp. 443453,
Apr. 1982.
[22] B. Sinopoli, L. Schenato, M. Franceschetti, K. Poolla, M. Jordan, and S.
Sastry, Kalman filtering with intermittent observations, IEEE Trans.
Autom. Control, vol. 49, pp. 14531464, Sep. 2004.
[23] S. Hu and W. Y. Yan, Stability of networked control systems subject
to input and output packet loss, in Proc. 48th IEEE Conf. Decision
Control, Held Jointly With the 28th Chinese Control Conf., Dec. 2009,
pp. 68886892.

[24] L. E. Ghaoui and M. A. Rami, Robust state-feedback stabilization of

jump linear systems via lmis, Int. J. Robust Nonlinear Control, vol. 6,
pp. 10151022, Nov. 1996.
[25] Y. Loparo and K. A. Fang, Stochastic stability of jump linear systems, IEEE Trans. Autom. Control, vol. 47, pp. 12041208, Jul. 2002.
[26] M. Chilali and P. Gahinet,
design with pole placement constraints: An lmi approach, IEEE Trans. Autom. Control, vol. 41, pp.
358367, Mar. 1996.
[27] D. Arzelier, J. Bernussou, and G. Garcia, Pole assignment of linear uncertain systems in a sector via a lyapunov-type approach, IEEE Trans.
Autom. Control, vol. 38, pp. 11281132, Jul. 1993.
[28] B. Pal and B. Chaudhuri, Robust Control in Power Systems. New
York: Springer, 2005.
[29] N. Martins and L. T. G. Lima, Determination of suitable locations for
power system stabilizers and static VAR compensators for damping
electromechanical oscillations in large scale power systems, IEEE
Trans. Power Syst., vol. 5, pp. 14551469, Nov. 1990.
[30] M. E. Aboul-Ela, Damping controller design for power system oscillations using global signals, IEEE Trans. Power Syst., vol. 11, no. 2,
pp. 767773, May 1996.
Abhinav Kumar Singh (S12) received the B.Tech.
degree from Indian Institute of Technology, Delhi,
India, in 2010 in electrical engineering (power).
Currently, he is pursing the Ph.D. degree at the
Imperial College London, London, U.K. His current
research interests include communication aspects of
power systems, dynamic state estimation and power
system dynamics.

Ravindra Singh (S07M09) received the B.Tech.

(with honors) degree from HBTI, Kanpur, India,
the M.Sc. (Engg.) degree from the Indian Institute
of Science, Bangalore, India, and the Ph.D. degree
from Imperial College London, U.K., in 2002, 2004,
and 2009, respectively, all in electrical engineering.
Currently he is working as a Research Scientist at the
ABB Corporate Research Center, Raleigh, NC, USA.
His current research interests include distribution
system state estimation and power system dynamics.

Bikash C. Pal (M00SM02F13) received the

B.E.E. (with honors) degree from Jadavpur University, Calcutta, India, the M.E. degree from the
Indian Institute of Science, Bangalore, India, and
the Ph.D. degree from Imperial College London,
London, U.K., in 1990, 1992, and 1999, respectively, all in electrical engineering. Currently, he
is a Professor in the Department of Electrical and
Electronic Engineering, Imperial College London.
He is Editor-in-Chief of IEEE TRANSACTIONS ON
SUSTAINABLE ENERGY and Fellow of IEEE for
his contribution to power system stability and control. His current research
interests include state estimation and power system dynamics.